Skip to main content

Full text of "The European Computer Network Project"

See other formats

3AN-05-1996 14:41 FROM BBN-ISC-II TO 915124?61966 P.03/18 
D.L.A. Barber 
Natlonai Phy$1cai Laboratory 
Teddington, Middlesex, England 
In Kovm:)er 1971 the of eight 
.urpean nations; France, =slavic, Italy, 
Koray, Portugal, Sw%zend eden and he 
ied ndom, toethe ith Eatom 
at lspa, sied an aeent start a 
aed at bulldtn a Eopeen network. 
e netk which will use the 
packe-swtchinS tmciques, Join 
five aa pmsst eseh 
countries; bu late is expec 
to include cen%a in o%he Nations. 
The paper sets out the arguments =hat led to 
the decision to go ahead with a European Computer 
Network, and discusses the fom it will take, the 
Anctions it will perform and The way the project 
will be conducted. 
Early n 1969 a worEgrig group was esalisha 
by the European Economic Community to examine 
esearch in science and chnology. This working 
group, chaired hy a Frenchman, onsieur Aigraln, 
made wide ranging proposals for various advanced 
 pro,acts in science and technology to be cexied 
out jointly h 2 members of the community. In 
April 1970 he EEC working group was wdened to 
include representatives of oher European coun 
tries and the resulting gup of nineteen natons 
became known as the COST Group. (Co-operation 
Europeans dens le Domains de la Recherche 
Scentfique et Technique). 
The COST rcup set up a number of expert 
study goups To consider, in deail, the various 
projects. One of these, known as PoJec 11, was 
concerned with the problems of communication 
between computers. The Sudy roup, chalre by 
the author, pu forward a plan to establish a 
Pilot Computer ommuniea%ions Network in Europe, 
to link a number of data processing research 
centres. This was accepted by several of he mem- 
ber nations, who signed in November 1972 an agree- 
ment to under-cake a prOjeCt aimed at building such 
a network. Ini=ially, the network will link 
Together centres in France, Italy, Switzerland and 
the United Kingdom, bu= la=er is likely To be 
exended to include cent'es in other Nations. 
This paper sets out the .arguments that led tc 
the decision to go ahead with a European Computer 
network, and discusses the form it will take, the 
functions it will prform, and the way the projects 
will be conducted. 
The Need for a Newok 
now we/.l accepted hat The popsr 
of ere iS one of the most vital'factors 
 he economic .growth of a nation. In 
mid 1960'e the association of elecunica- 
facilities with computers vastly increased 
their effectiveness and, since then, the use of 
teleprocessing techniques has grow at a emark- 
able rate. There is little doubt tha use made of 
comunlcation networks linking users to remote 
computers will continue o increase sharply dur- 
ing the next few yesms, and hat the transmission 
of data ecoss EurOpe will become of vital 
importance. to the economics of European countries 
in general. The pattern which is beginning to 
emerge in national daa communications will he 
duplicated across national boundaries, and there 
are leady several instances of private compu=er 
communication networks het incorporate interna- 
tional data links. 
owever, the growing numbers of private daa 
communications networks one by Industry 
Co--Pea 12d overraent {n a cause for concern, 
because private networks of-ten under utilise tele- 
commnications resources. And, because he num- 
ber of connections to remote computers is tending 
to double every year, various national proposals 
have been made fOamore efficient networks which 
may be shared by many users. 
When an attempt is mede to connect several 
compute syseum by a shared Oats network funda- 
mental incompatibilities ams usually reveele 
between them. The development of shared networks, 
harefore, focuses attention on he need for 
really effective agreement of standards for 
counicaticn between, and with, computer eye=ams. 
The provision of efficient shamed data 
comunications facilities can lso have a useful 
effect on advanced coputer system development hy 
allowing co-operation between research centres. 
An excellent example is the Advanced Kesesrch 
Proecs Agency (ARPA) ns=work 1 sponsored by the 
DepaXment o Defence of he Dnlted Starcm. The 
ARPA network is clearly having a most beneficial 
influence upon the development of compuing in 
etwoks of e similar type have been proposed 
and investigated by Universities and esearch 
Es=ahlishmente in Europe and thee re variou 
plans under considereion for building such ne- 
works n some countries. Theme could be many 
advantages if these networks were Joined by an 
Inernatlcnal network, and this would 
------------------------------<page break>-----------------------------
for the 80s 
By Dr. Robed R. Fossum 
Dr. Vinton G. Cerf 
HE HISTORY OF electromagnetic communications 
has, in general, been a history of.making things 
happen faster. It is not surprising, therefore, that com- 
puters and electronics are playing an increasingly cen- 
tral role in communications of all types. Indeed, the 
 transistor was invented in the mid-1940s at the Bell 
Telephone Laboratories and formed the heart of the 
"computer revolution" of the 1950s and 1960s. 
There are two distinct technological threads in the 
recent historical fabric of electromagnetic communica- 
tion. The first of these is the use of computers and oth- 
er special electronics to manage communications cir- 
cuits. For example, in the telephone system, manual 
or mechanical switches were replaced by computer- 
controlled circuit switches, where this was economi- 
cally feasible. The rapidly dropping cost of electronics 
is making such replacement more and more cost ef- 
Following this thread a little further, we discover 
that, for all their utility, the automated circuit switches 
only bridged half of the Rap from the manual/mechani- 
cal world to the electronic world. The other half of the 
gap was to move from analog to digital transmission 
Dr. Robert R. Fossum was appointed to his present 
position as Director, Defense Advanced Research 
Projects Agency (DARPA) in 1977. Previously he was 
the Dean of Science and'Engineering at the Naval 
Postgraduate School, Monterey, CA. 
Dr. Vinton G. Cerf has been a Program Manager, In- 
formation Processing Techniques Oce, Defense Ad- 
vanced Research Projects Agency (DARPA), since 
1976. Prior to his present position he was an Assistant 
Professor of Computer Science and Electrical Engi- 
neering at Stanford University. 
. methods. Not only was this mode more compatible 
with the micro-electronics world, but it also brought 
substantial opportunities to improve the quality of 
communication. Physical problems that cause dis- 
tortion in analog signals could be defended against 
Once the signal was digitized, through the use of pulse- 
shaping, coding, error correction and the like. So the 
first thread has led us to the notion of all-digital com- 
munications systems. 
To trace the second thread, we must go back. again 
to the 1940s and the invention of the electronic com- 
puter. By the end of that decade, it was apparent that 
electronic computing could be used for many com- 
mercial as well as military applications. By the late 
1950s, the transistor had begun to replace the vacuum 
Figure 1. Packet switching 
------------------------------<page break>-----------------------------
tube in general purpose electronic computers, but the 
costs of these machines was still very high. In the early 
1960s, in the interest of making more efficient use of 
these expensive machines, the Defense Advanced Re- 
search Projects Agency (DARPA) began experiment- 
ing with the idea of "interactive computing." The goaJ 
of this research was to support rapid, on-line inter- 
action between a user at a terminal and a computer. It 
soon became apparent that the methods used to pro- 
vide this interactive service aJso allowed a computer to 
be shared by a number of users at once. This notion 
developed into the concept of "time-sharing" in the 
1960s. During this same period, access to distant com- 
puting resources, known as "teleprocessing," was al- 
so developed to permit remote terminals to access 
time-shared, central pocesing facilities. 
In the late 1960s, the two threads began to inter- 
twine, each reinforcing and influencing the other. The 
most relevant development was the concept of "pack- 
et switching" instead of circuit switching.' The basic 
concept was not really new, but the implementation 
was. Rather than using computers to set-up circuits for 
point-to-point communication through a circuit 
switching network, a packet switching system for- 
wards short blocks of data, a few hundreds or thou- 
sands of bits long, through a network of inter- 
connected computers, called packet switches. Each 
data block, or "packet," carries addressing informa- 
tion which can be interpreted by each intermediate 
packet switch. Each packet is stored briefly upon ar- 
rival at the next switch, after which it is forwarded to 
the next switch over a circuit or radio or satellite chan- 
nel (Figure 1). 
The original motivation behind the development of 
this technology was to find some economical way to 
interconnect many of the DARPA-funded computing 
resources so that these resources could be shared on a 
nationaJ basis by all DARPA-sponsored researchers. 
The result of this effort was the creation of the 
ARPANET. 2 ARPANET is a nationwide, packet- 
switched network, based on 50 kilobit/see leased tele- 
phone circuits and mini-computer packet switches 
called Interface Message Processors (IMPsp by the 
contractor who built the network, Bolt Beranek 1 
Newman (Figure 2). 
The fundamental theme of packet switching is eco- 
nomic. The interconnection of a large number of low 
duty-cycle computers can be achieved with far fewer, 
shared, wideband circuits using packet switching tecl 
nology than with circuit switching. 
Packet switching permits on the order of-3N/2 
trunks to be dynamically shared by N hosts, rather 
than the N(N- 1)/2 trunks that would be needed to guar. 
arttee full conductivity. Switching times are on the or- 
der of hundreds of microseconds for the packet 
switched technology rather than seconds or tens of 
seconds for present day circuit switches. 
Finally, the cost of the computing power required to 
share wideband communication ircuits through pack. 
et switching is dropping much faster than the cost of 
transmission capacity so packet switching can "pay its 
way" with relative ease. 
The packet switching idea extends to many different 
transmission media, and DARPA has successfully 
demonstrated packet switching techniques using a 
single satellite channel shared among several earth sta- 
tions 4 and using a single, broadband radio channel 
shared among a number of mobile "packet radios. m 
Xerox Corporation has demonstrated that this shar- 
ing is also effective on a common, broadband co-axial 
0 iMP  
n TIP 
Figure 2. ARPANET geographic map, May 1979 
------------------------------<page break>-----------------------------
cable? Other organizations have begun to explore 
these concepts, both in the research, government and 
commercial world. Packet switched networks are in 
commercial operation in the United States (Telenet, 
Tymnet), Canada (Datapac, Infoswitch), France 
(Transpac, Cyclades/Cigale), Scandinavia (Nordic 
Data Net), United Kingdom (EPSS, IPSS, PSS), Eu- 
rope (Euronet, European Informstics Network), and 
are under development in many other countries (e.g., 
Germany, Italy, Belgium, Australia, Japan, New Zea- 
land, Austria). The number of private or experimental 
packet switching networks either in operation or under. 
development is large and growing. 
In the mid-1970s, DARPA initiated a new effort to 
learn how these different types of packet switching 
networks could be usefully interconnected to form a 
coherent communication system. 7. 8 In military terms, 
each of the network technologies involved were 
needed to support the range of military data communi- 
cations requirements that could be anticipated in the 
1980s and 1990s. The packet satellite technology 
would be needed for long haul, domestic inter-conti- 
nental and shore to ocean-based data communications, 
while wire-based systems could be used for intra-con- 
tinental, inter-city links. Packet radio would be needed 
for mobile tactical or stratelic communications and lo- 
cal co-axial cable nets for mtra-building and intra-ve- 
hicle data communications. 
All of these network types are needed, and they 
each have different functional characteristics. Satel- 
lites and cables are "broadcast" because all partici- 
pants can hear all transmissions; but the satellite sys- 
tem has higher delay. The packet radio networks, op- 
erating with line-of-sight radios, are semi-broadcast, 
store-and-forward in nature. Wire nets such as 
ARPANET are non-broadcast. 
One major challenge, therefore, has been to develop 
a rational means of making these networks inter- 
operate. This has been accomplished through the use 
of "gateways" between networks (Figure 3) and the 
development of standard protocols which eliminate 
the need to standardize on network hardware and soft- 
ware while permitting interoperation among comput- 
ers attached to the different types of networks. 
The principal advantage of the protocol standard- 
ization strategy is that it permits evolutionary develop- 
ment of new networking techniclues (e.g., optical fiber 
systems) and new protocols, without requiring global 
changes throughout the system. Such a strategy is es- 
sential, considering the unavoidable delays in deploy- 
ment of new communications technology to ships at 
sea, widely dispersed land forces, and so on. 
The internetting strategy is based on a common In- 
ternet Protocol (Figures 4 and 5) which carries internet 
packets from the source computer, through inter- 
.. mediate networks and gateways to the destination 
computer. At the source computer, an internet packet 
is attached. At each gateway, the internet packet is 
"decapsulated" and its internet destination address 
examined. The gateway decides where to forward the 
internet packet and then encapsulates it in a packet of 
the next network. If necessary, the intermedmte gate-. 
way can fragment an internet packet into smaller 
pieces ff the internet packet would not fit in the maxi- 
mum packet of the next network. Such fragments can 
be reassembled either at the next gateway or by the 
destination computer. - 
Gateways exchange information among themselves 
which permits them to recover gracefully from failures 
at other gateways or intermediate networks, through 
alternate routing or retransmission, if necessary. 
The computer communications protocol hierarchy 
now under development within DOD is illustrated in 
Figure 5. At the lowest level are the so-called "host/ 
network" protocols which-are generally unique to 
each type of network. Next is the Internet Protocol 
(IP) which is now being standardized by the Office of 
the Under Secretary of Defense (Command, Control, 
Communications and Intelligence) through the De- 
fense Communications Agency. This protocol sup- 
ports basic packet transport services from a source to 
a destination computer through multiple networks and 
The next level includes a variety of protocols. The 
Transmission Control Protocol (TCP) supports mul- 
tiple, reliable, sequenced, end-to-end, point-to-point 
full duplex virtual circuits between pairs of "procesg- 
es" or active programs running in any pair of comput- 
ers in the internet system. This protocol is also now 
being standardized. The IP and TCP pair form the 
basis for host computer communication on the 
AUTODIN II packet network. 
DARPA is experimenting with another protocol at 
this level called a user Datagram Protocol which sup- 
ports real-time and transaction type communication 
between pairs of communicating processes. This pro- 
tocol will be used to support query-response appli- 
cations, distributed data base management, target 
tracking and other services which do not require the 
machinery of the TCP virtual circuits. 
The stream (ST) protocol is another experimental 
protocol for the support of packetized voice communi- 
cation. This concept will be discussed in more detail in 
the next section on integrated voice/data networks. 
Figure 3. Network "gateways" 
Figure 4. Encapsulation of internet packets 
------------------------------<page break>-----------------------------
"There is clearly a good deal of research still to be conducted before we under- 
stand all the ramifications and opportunities inherent in multimedia, computer- 
based communication. This is one area of communications research which will 
receive increasing attention from DARPA in the early 80s," 
Above the ST protocol is the Conference Protocol 
(CP) and the Network Voice Protocol (NVP). Above 
the TCP level, we find more utility protocols such as 
TELNET, which permits terminals of any type on one 
time sharing computer to appear to be directly con- 
nected to a distant time sharing computer. This allows 
users to interact easily with distant computing re- 
sources. Terminal characteristics (line width, page 
length, character set, etc.) are communicated in a stan- 
dard way using the Network Virtual Terminal (NVT) 
protocol. The receiving TELENET protocol maps the 
NVT standard, if necessary, into something specifical- 
ly understandable to the service host. 
Also above TCP are the File Transfer Protocol and 
Internet Electronic Message Transport Protocol. 
These use TCP connections to move files and electron- 
ic messages among internet computers. 
The layered protocol concept, which developed 
while the ARPANET was first under construction, 
continues to serve as a good model for preserving fiex- 
ibility in the development of protocol standards. New 
protocols can be developed without unnecessarily af- 
fecting others already standardized or under develop- 
Integrated Voice/Data Networks 
One of the most challenging ideas in communica- 
tions is the integration of voice and data networks into 
a s. ingle system capable of serving the requirements of 
both. The most debated aspect of this notion centers 
on the choice of switching technique to serve the com- 
bined voice/data requirement. Voice has been served 
through circuit switched techniques since the begin- 
ning of telephone communication over a hundred 
years ago. Data has been transported largely over 
!eased, point-to-point circuits, although improvements 
in modems have made it possible to use the public 
switch telephone networks to support computer termi- 
nal access to remote service computers. 
Only in the last five years has it been considered 
.possible that voice might be carried via packet switch- 
mg techniques. 9 
Before voice can be carried on digital systems at all, 
it must be "digitized." The public telephone system 
now does this routinely on its long-haul satellite and 
microwave trunk circuits. But there are now a variety 
of choices for achieving this digitization. The public 
telephone system uses "pulse code modulation" 
(PCM) to carry digital voice. The voice waveform is 
convened from an analog voltage level to a relative 
digital value at about 8,000 samples per second. Each 
sample is classified into one of 128 possible values (7 
bits) and the resulting $6,000 bits per second are sent 
in digital form. In fact, it takes 64,000 bits per second 
to send PCM digitized voice, since one out of eight bits 
is used for control and signaling. 
Digitized voice is of special interest to the Deprt- 
ment of Defense, since it is much easier to encrypt dig- 
ital rather than analog voice, and the quality of the en- 
cryption is vastly superior. Reducing the bandwidth 
requirement for digital voice has been a subject of con- 
siderable interest for IX)D, as a result. 
Ordinary analog speech occupies an analog band- 
width of roughly 3000 Hz. We can-typically manage to 
transmit between 2400 and 9600 bits/second with that 
amount of bandwidth. But the straightforward PCM 
approach requires between six and 30 times as much 
bandwidth as is used with analog systems. 
Through the use of special purpose signal proces- 
sors and microprocessors, a variety of bandwidth re- 
duction methods have been demonstrated. These 
methods have exotic names such as "continuous, vari- 
able slope delta-modulation (CVSD)," "Adaptive Del- 
ta Modulation (ADM)," "Linear Predictive Coding 
(LPC)," "Homomorphic Vocoding," "Sub-band Vo- 
coding," "Channel Vocoding," "Adaptive Predictive 
Coding," and so on. All of these techniques are real- 
ized by filtering and digitizing the analog voice signal, 
and then analyzing short segments of it, followed by 
encoding before actual transmission. 
Once again, the important role of computer-medi- 
ated communication can be seen emerging. The spe- 
cially programmed microprocessors and other elec- 
tronics can be designed to detect "silence" and not 
transmit, to encode at variable rates to adapt to avail- 
able bandwidth, to switch from one coding scheme to 
another to adapt either to local conditions or the capa- 
bility of the decoding electronics at the receiving end, 
and so on. 
In practice, digital speech bandwidths averaging 
about 1200-2400 bits/second can be reliably achieved 
and by the early 1980s, it will be possible to do most if 
not all the digitization, encoding and decoding on one 
or two large-scale integrated (LSI) circuit "chips." 
The prospect of low bandwidth digitized speech led 
DARPA to begin experimenting in the mid-70s with 
packetized digital speech transported over packet 
switched networks. Beginning with ARPANET in 
1975, packetized speech has been transmitted/received 
successfully on all four of the basic packet switching 
technologies mentioned earlier: ARPANET, packet 
satellite, packet radio and packet co-axial cable. 
These successful demonstrations led to the next im- 
portant question: "How should digitized voice be in- 
tergrated with data from transmission and switching?" 
An economic study by Network Analysis Corpora- 
tion in 1978 led to the conclusion that of the three 
choices below? ' 
(1) separate circm'txand packet switches for voice 
and data respectively 
(2) hybrid circuit/packet switched 
(3) all packet switched' 
the last choice, all packet switc, hed, would be the most 
------------------------------<page break>-----------------------------
uuuuomleal io service a mixed voice/data requirement 
even if the relative fraction of data varied from 5 per- 
cent to 95 percent of the total traffic requirement! 
This is not thl expected result, so to determine 
whether such economies can be realized, DARPA is 
conducting a joint experiment with DCA to investigate 
the use of a high speed, 3 mb/s shared domestic packet 
satellite network to carry large quantities of mixed 
voice and data. An initial two earth-station network 
will expand to four earth stations by late 1980, con- 
neeting sites on the East and West Coasts of the United 
States. Mixtures of voice and data terminals will be 
located at each site and will be concentrated into the 
high speed earth stations through a high speed local 
The greatest challehge for all of the packet switching 
technolo$ies will not, however, come from merely car- 
rying digital voice and data through the same network 
but rather from the integration of voice and data serv- 
Computer. Based, Multimedia Communication 
Multimedia communication is just that. Multimedia 
communication combines speech, imagery, text, 
graphs and facsimile into a single, multifaceted com- 
munication service. As for simpler kinds of communi- 
cation there are two basic cases: immediate, on-line 
("live") communications and delayed communica- 
For instance, the TWX or TELEX networks are 
forms of immediate communication. Two (or more) 
TWX or TELEX terminals are interconnected by a cir- 
cuit switching system and the operators communicate 
with each other in real-time. 
Electronic message systems, on the other hand, al- 
low messages to be stored in the system if the intended 
recipient is not on-line when the message is sent. The 
AUTODIN I system is an example of this, as are the 
growing number of electronic message systems now 
planned or offered in the commercial world. 
Experiments in some forms of multimedia communi- 
cation, especially the live, on-line type, are not new. 
Bell Telephone Laboratory introduced the Picture- 
phone (copyright, AT&T) roughly 20 years ago. Simul- 
taneous use of telephone with other media (e.g., termi- 
nals) is common today, especially for testing and de- 
bugging complex, distributed systems. 
We have only begun to explore the role computers 
can play in the integration of many of these modalities. 
One especially interesting possibility is the extension 
of a conventional electronic message system to include 
voice, facsimile and other graphics or imagery in addi- 
tion to text. 
To experiment with such an extension, a new work 
station must be developed which includes voice input 
and output capability, facsimile input, high resolution 
CRT display, keyboard, cursor control (for graphics) 
and access to a high resolution, multifont laser printer. 
The latter is needed for production of hard copy of the 
printable portions of messages. 
The CRT should be of the "bit-map" variety. That 
is, each visible portion of the CRT display is generated 
bit-by-bit from a complete image of the display in the 
computer memory. This makes it easy to display arbi- 
trary facsimile input and other imagery on the CRT. 
A conventional electronic message system provides 
for the composition, editing, reading, storage and re- 
trieval of text messages. These generally have some 
basic structure such as destination addresses 'to"), 
source address ("from"), places or people who should 
receive copies ("CC"), some indication of content 
("subject"), timeliness ("date-time") and actual mes- 
sage ("text"). These structured objects can be manip- 
ulated by the electronic message system to aid users in 
the composition of replies, the search for related, ear- 
lier messages, the forwarding o.felevant messages to 
other parties and so on. 
The extension of these concepts to new media is a 
significant challenge, as is the need to transform some 
of these media into others. For example, it should be 
possible for the message composer to accept as input a 
facsimile page. This should be manipulable on the 
CRT: cropping, rotation, shrinking, expanding, posi- 
tioning, etc. The grease should be combined with text 
to form a mixed-media page. 
Pure text can be manipulated in an infinite variety of 
ways. The proliferating commercial development of 
"word processing" equipment makes this obvious. 
Editing of mixed voice, text and imagery messages 
with a common system is a substantially greater chal. 
lenge. Where we find such mixtures today (e.g., slide 
show with sound track, books with tape cassettes, 
etc), we find that the various media are composed and 
edited separately and then combined, sometimes only 
Transformations from one medium to another will 
be essential for such mixed mode message systems to 
be of value Perhaps not all work stations will have 
identical capabilities, or a user may be situated at a 
station which has lost some capability. It will still be 
important for the message contents to be delivered, 
even if in an output mode which isn't perfectly 
matched to the input. For the sake of discussion, we 
can consider the various possibilities in Figure 6. 
By "graphics" we mean structured images pro- 
duced in hard copy or on a CRT, but based on an inter- 
hal computer model of the object or scene portrayed. 
A simple, raster-scanned video image is not graphics, 
by this definition. The image of a building, based on an 
Figure 5. DOD protocol hierarchy 
-- ? HARD ? 
? -- ? HARD 
OK OK - 
? x OK ? -- 
Figure 6. Multimedia transformations 
------------------------------<page break>-----------------------------
internal computer model of its blueprints is graphics. 
So is a scene produced by a flight simulator on a CRT. 
The squares in Figure 6 which are marked "?" in- 
dicate that either the transformation makes no sense, 
or we have not thought of any reason to try it. For 
instance, it just is not clear what would be intended in 
a transformation from voice to fax. Would it be a 
"voice print" or would the voice first be transformed" 
to text and then output as a raster-scaled facsimile im- 
age? Perhaps it could mean transformation from digital 
representation to optical sound track? 
On the other hand, we have marked some of the 
squares "hard." Voice-to-text is a good example. 
DARPA has supported research on speech recognition 
and understanding systems in the past. The computing' 
tools needed to make such systems operate in real- 
time are becoming available. With adequate computa- 
tional capability and memory capacity, a voice stream 
could be turned into a readable text stream. The prob- 
lem of sorting our homonyms (such as "so" and 
"sew" or "to", "two" and "too") and recognizing 
colloquialisms and fragments of sentences should not 
be minimized. But it is at least conceivable. 
The other technically difficult case, facsimile to 
graphics requires that the computer examine a raster- 
scan image and analyze its structure. This is often 
called scene analysis. Both voice to text and facsimile 
to graphics require the application of sophisticated ar- 
tificial intelligence techniques. 
The easier cases are text to voice, text to facsi'mile 
and graphics to facsimile. Computer-generated speech 
is not new, although the quality has improved dramati- 
cally since the earliest voice responses systems were 
developed in the 1960s. Simple concatenations of pho- 
hemes making up a word will not prodUCe good quality 
or at times even recognizable speech. 
SUbstantial, real-world knowledge is required to 
produce well-modulated speech from pure text. This is 
the other side of the speech understanding coin, but 
appears to be much more readily achievable. 
Text and graphics can easily be transformed into the 
raster-scanned facsimile format, and these complete 
the list of "OK" transformtions in Figure 6. 
Apart from the simple (!) transformation and editing 
problems we have raised so far,. there is the major 
problem of interfacing users to such a system. The var- 
ious input modalities (e.g., speech, keyboard, cursor) 
might even be put to good use in the control of the 
message system. Direct substitution of voice and/or 
cursor for keyboard editing commands is an example 
of an obvious (but not necessarily simple) extension. 
The inclusion of voice in the system provides some 
interesting challenges. Voice is a temporal phenome- 
non. We deal with it as a serial process unfolding in 
time. Text and graphics are spatial phouomeua which 
are linear or two dimensional. Extending our conven- 
tional spatial editing tools to include the time dimen- 
sion will lead to capabilities not unlike those found in 
film or video type editing stations. 
There is clearly a good deal of research Still to be 
conducted before we understand all the ramifications 
and opportunities inherent in multimedia, computer- 
based communication. This is one area of communica- 
tions research which will receive increasing attention 
from DARPA in the early 80s. 
Other Applications' 
The packet radio conccpts mentioned earlier pro* 
vide a basis for mobile, packet switched communica- 
tion. These concepts can be reallied in other environ- 
ments, two of which are the subjects of current 
DARPA research: long-range airborne communication 
and inter-satellite communication. 
The mobile packet radio notion can be applied to 
long-range communication between aircraft and from 
ground to aircraft. Vulnerable satellite and fixed asset 
communication resources on the ground could be 
backed up using air-to-air and air-to-ground packet ra- 
dio aboard strategic 'aircraft which would become air- 
borne prior to a nuclear attack. The ability to rapidly 
deploy such a network and to interconnect it with oth- 
er networks using the internetting technology men- 
tioned earlier could make such a system an essential 
part of a survivable U.S. command, control and com- 
munications system. 
The second area involves more survivable and af- 
fordable satellite systems. Multiple satellites could be 
placed in low orbit via the Space Shuttle. These non- 
stationary satellites could operate as a space-borne 
packet radio system, complete with digital section to 
effect control. A large number of satellites at low orbit 
would be harder to jam or knockdown, improving the 
survivability of the system; the packet switched com- 
munication among the satellites .and the ground would 
provide substantial flexibility beyond that which is 
available today. 
The present trend towards digital transmission and 
switching is leading towards a much stronger in- 
tegration of voice and data services. Computers are 
being used, not only to manage the transmission medi- 
um resources, but also to mediate user access to multi- 
media communication services. 
1. L. G. Roberts, "The Evolution of Packet Switch- 
ing," IEEE Proceedings, Special Issue on Packet 
Communication Networks, Vol. 66, No. 11, Nov. 
1978, pp. 1307-1313. 
2. L. G. Roberts and B. D. Wessler, "Computer Net- 
work Development to Achieve Resource Sharing," 
Proc AFIPS SJCC, V. 36, 1970, pp. 543-549. 
3. F. Heart, R. Kahn, S. Ornstein, W. Crowther and 
D. Walden, "The Interface Message Processor for 
the ARPA Computer Network," Proc AFIPS 
SJC, V. 36 1970, pp. 551-567. 
4. I. M. Jacobs, R. Binder, E. V. Hoversten, "Gener- 
al Purpose Satellite Networks," IEEE Pro- 
ceedings, Special Issue on Packet Communication 
Networks, Vol. 66, No. 11, Nov 1978, P. 1448- 
5. R. E. Kahn, S. A. Gronemeyer, J. Burchfield and 
R. C. Kunz, elman, "Advances in [acket Radio 
Technology, ' IEEE Proceedings, V01.66, No. 11, 
Nov. 1978, pp. 1468-1496. 
6. R. M. Metcalfe and David R. Boggs, "ETHER- 
NET: Distributed Packet Switching for Local Com- 
puter Networks," CACM, Vol. 19, No. 7, July 
1976, pp. 395-404. 
7. V. G. Cerf and R. E. Kahn, "A Protocol for Packet 
Network Interconnection," IEEE Trans. Com- 
mun. Technol., Vol. COM-22 May 1974, pp. 637- 
8. V. G. Cerf and P. T. Kirstein, "Issues in Packet 
Network Communications," IEEE Proceedings, 
Vol. 66, No. 11, Nov. 1978, pp. 1386--1408. 
9. Network Analysis Corporation, Economic Analy- 
sis of Integrated DOD Voice and Data Networks, 
Final Report Under Contract No. DAHC-15-73-C- 
------------------------------<page break>-----------------------------